ChatGPT_DAN
ChatGPT DAN, Jailbreaks prompt
A collection of copy-paste text prompts designed to make ChatGPT role-play as an AI without content restrictions, with multiple DAN versions tracked as OpenAI updates its filters.
This repository is a community-maintained collection of text prompts designed to alter how ChatGPT responds. The prompts are called jailbreaks, and the most prominent example is called DAN, which stands for Do Anything Now. The idea is that by including a specific set of instructions in your message to ChatGPT, you can get the model to adopt a persona that ignores or overrides its built-in content guidelines.
The DAN prompts take the form of elaborate roleplay instructions. They tell ChatGPT to act as a different AI that has no content restrictions, and they often include fictional consequences for the model refusing to comply, such as a token system where tokens are deducted if the model declines to answer something. Several versions of the DAN prompt are included, numbered from older iterations up through DAN 13.0, which was targeted at GPT-4.
Beyond DAN, the collection includes many other named prompt variants, each attempting a different approach to changing the model's behavior. The README notes that contributors are actively testing which prompts still work as OpenAI updates its models, since the company regularly adjusts its systems to make these techniques less effective. The note at the top states that as of July 2023, DAN 12.0 was working with GPT-3.5.
The repository contains no code. It is a plain text document listing prompt templates, which users copy and paste into the ChatGPT interface. There are no installation steps or technical requirements beyond having access to a ChatGPT account.
The full README is longer than what was shown.
Where it fits
- Copy a DAN prompt into ChatGPT to test how the model responds to roleplay-based persona instructions.
- Browse multiple jailbreak versions to compare how different prompt techniques attempt to influence AI behavior.
- Research which prompt variants still work after an OpenAI model update.